Claude Haiku 5.5, which Anthropic released on October 7, 2026, is the first Haiku worth picking on purpose, and Anthropic says every Claude plan can use it on Claude.ai, including Free. It is the third model in the Claude 5.5 family, after Opus 5.5 and Sonnet 5.5. Developers pay $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens. Anthropic says that makes it about 75% cheaper to run than Claude Haiku 4.5.
Most coverage of the launch is written for developers: the API price, the price step above 100,000 tokens, the race with OpenAI's GPT-6 Luna. This guide covers that, and then answers the question a Claude user actually has. When Haiku 5.5 shows up in your model menu, when should you pick it over Sonnet 5.5, what does it save you, and what is it bad at?
The Key Takeaways
- Free on Claude.ai: Anthropic says Free, Pro, Max, Team and Enterprise users can select Haiku 5.5 on the web, iOS and Android.
- Lightest on your limit: Anthropic's own guide calls Haiku the lightest Claude model on your usage limit, so it stretches a Free or Pro allowance furthest.
- A year's jump: Artificial Analysis scores Haiku 5.5 at 43 on its Intelligence Index, up from 17 for Haiku 4.5 and ahead of GPT-6 Luna at 38.
- Same price as GPT-6 Luna: $0.10 / $0.50 per million tokens up to 100,000 tokens, rising to $0.50 / $2.50 above that.
- Knows less, guesses less: it answered fewer factual questions correctly than Gemini 3.8 Flash or GPT-6 Luna, but its hallucination rate was the lowest of the three, at 40%.
What Claude Haiku 5.5 Is
Claude Haiku 5.5 is Anthropic's smallest, fastest and cheapest current model. Anthropic calls it "the cheapest, fastest, and most capable small model we've ever released" and built it for quick, repetitive work: summaries, classification, database queries and short lookups. It replaces Claude Haiku 4.5, which launched on October 15, 2025, so the two models are less than a year apart.
Anthropic announced the model on its Claude account:
Introducing Claude Haiku 5.5: the cheapest, fastest, and most capable small model we’ve ever released.
— Claude (@claudeai) October 7, 2026
On average, it costs around 75% less to run than Claude Haiku 4.5.
The specs that matter
Haiku 5.5 has a 1 million token context window and returns up to 128,000 output tokens, according to Anthropic's developer docs. Haiku 4.5 stopped at 200,000 and 64,000. Haiku 5.5's knowledge cutoff is June 2026, and it is the first Haiku with an adjustable effort setting, the same control Sonnet 5.5 and Opus 5.5 already had. Higher effort means more thinking, better answers and a bigger bill. The default is medium.
| Spec | Haiku 5.5 | Haiku 4.5 |
|---|---|---|
| Released | October 7, 2026 | October 15, 2025 |
| Context window | 1M tokens | 200K tokens |
| Max output | 128K tokens | 64K tokens |
| Effort setting | Yes, default medium | No |
| API price per 1M tokens | $0.10 / $0.50 up to 100K tokens | $1 / $5 |
Where it sits in the Claude lineup
Haiku 5.5 is the bottom of four current tiers. Claude Fable 5.1 is the top model, Anthropic calls Claude Opus 5.5 the daily driver for complex coding and knowledge work, and Sonnet 5.5 the fit for well-scoped tasks. Anthropic's pitch is that the tiers work as a team: a big model plans, and Haiku does the high-volume legwork. Mike Krieger, Anthropic's former chief product officer who now works at Anthropic Labs, put it in one line:
Another model joins the 5.5 family with the small but mighty Haiku 5.5! It’s built for speed and efficiency, and costs 75% less than Haiku 4.5. Opus does the heavy thinking, Haiku does the high-volume work, and together you get more done for less. Happy shipping!
— Mike Krieger (@mikeyk) October 7, 2026
Is Claude Haiku 5.5 Free?
Yes, Claude Haiku 5.5 is free inside Claude. Anthropic's Haiku product page says "Free, Pro, Max, Team, and Enterprise users can select Haiku 5.5 on Claude.ai, available on web, iOS, and Android." Claude's pricing page lists Haiku on every plan, Free included. Opus starts at Pro. Developers who call the model through the API pay per token instead.
Anthropic does not say Haiku 5.5 is the default on the Free plan, so do not expect it to answer unless you choose it. Our guide to what Claude's free plan includes covers the message allowance and the other models Free accounts get.
Why Haiku stretches your usage limit
Haiku 5.5 uses the least of your usage limit of any Claude model. Anthropic's Claude Academy guide to choosing a model says the models "consume tokens at different rates: Haiku is the lightest, Sonnet is moderate, Opus is heavy, and Fable uses the most." Send a Haiku-sized task to Opus or Fable, the guide warns, and you use more of your limit for no gain.
That matters most on the Free plan, where Claude's allowance is small. Anthropic lists model choice and effort level among the factors that decide how fast you hit the cap, alongside message length, file size and tools like web search. It does not publish how many more Haiku messages you get than Sonnet messages, so treat that ranking as a direction, not a number. Our guide to how Claude usage limits work explains the five-hour and weekly meters. Another way round the cap is Fello AI, the Mac, iPhone and iPad app that puts Claude, GPT, Gemini and other models in one chat window. It has its own free tier, separate from your Claude account.
How to use Haiku 5.5 in Claude
- Open Claude on the web, or the Claude app on iPhone or Android.
- Open the model picker and choose Haiku 5.5.
- Leave the effort at its default for most work. Anthropic's model guide says lowering effort gives quicker answers that use less of your limit.
Anthropic lists the web, iOS and Android as the places to select Haiku 5.5. Its launch materials do not mention the Claude desktop apps for Mac or Windows either way.
Claude Haiku 5.5 Benchmarks: A Year's Jump
Claude Haiku 5.5 beats Haiku 4.5 on every benchmark Anthropic published, often by a wide margin, and it beats GPT-6 Luna on every test where Anthropic lists both. The biggest jump is computer use: on OSWorld 2.1, which tests whether a model can operate a real computer through long tasks, Haiku went from 15.7% to 72.4%. These are Anthropic's own numbers.
| Benchmark | Haiku 5.5 | Haiku 4.5 | GPT-6 Luna | Sonnet 5.5 |
|---|---|---|---|---|
| GDPval-AA v2.1 (knowledge work, Elo) | 1620 | 735 | 1437 | 1840 |
| AA-Briefcase v1.1 (knowledge work, Elo) | 1578 | 614 | 1336 | 1824 |
| OSWorld 2.1 (computer use, offline subset) | 72.4% | 15.7% | 48.9% | 83.9% |
| Humanity's Last Exam (no tools) | 45.9% | 10.2% | not listed | 56.9% |
| Terminal-Bench 4.0 (agentic coding) | 39.2% | 0.0% | 16.4% | 70.6% |
| FrontierCode 1.1 (agentic coding) | 46.4% | not listed | 42.4% | 52.1% (xhigh) |
| Chartography (visual reasoning, no tools) | 46.4% | 6.4% | 29.1% | 61.6% |
Sonnet 5.5 still leads Haiku 5.5 on every row, and by the widest margin on Terminal-Bench 4.0, 70.6% against 39.2%. Anthropic says so directly: Sonnet 5.5 and Opus 5.5 "remain better choices for complex agentic coding tasks". If you want the background on what each test measures, our guide to AI benchmarks decodes them.
The independent score
Artificial Analysis, which tests models independently, scores Haiku 5.5 at 43 on version 4.3.2 of its Intelligence Index, measured at maximum effort. Haiku 4.5 scores 17 on the same index. That puts Haiku 5.5 slightly ahead of other small models: GLM-5.3 Flash at 42, Gemini 3.8 Flash at 41 and GPT-6 Luna at 38. It sits level with Kimi K3 at 44, a much larger open-weight model, and 13 points behind Sonnet 5.5 at 56.
Anthropic has released Claude Haiku 5.5, scoring 43 on the Artificial Analysis Intelligence Index - up 26 points one year after the last Haiku release
— Artificial Analysis (@ArtificialAnlys) October 7, 2026
Haiku 5.5 is the first Haiku model with Anthropic’s effort settings and adaptive thinking, and Anthropic has introduced tiered pricing.
Artificial Analysis's own Terminal-Bench 4.0 run gave Haiku 5.5 33%, lower than the 39.2% Anthropic reports. That still puts it level with GLM-5.3 Flash and ahead of Gemini 3.8 Flash at 20% and GPT-6 Luna at 13%. The full results are on the Artificial Analysis model page.
Where Haiku 5.5 is weaker
Claude Haiku 5.5 knows less than its rivals but makes fewer things up. On AA-Omniscience, Artificial Analysis's factual knowledge test, it answered 36% of questions correctly, against 55% for Gemini 3.8 Flash and 44% for GPT-6 Luna. But it was more willing to say it did not know, and its hallucination rate was 40%, against 55% for Gemini 3.8 Flash and 77% for GPT-6 Luna.
Two more caveats from the same evaluation. Haiku 5.5 scored only 35% on AutomationBench-AA, against 53% to 60% for its rivals, because a safety refusal issue made it refuse too often in pre-release testing. Anthropic is working on a fix and Artificial Analysis expects the score to rise. And at maximum effort Haiku 5.5 is wordy: it used about 162,000 output tokens per index task, roughly three times GPT-6 Luna.
The practical rule: for trivia, use a model with web search. For summaries and rewrites of text you supply, the low hallucination rate is the better number to watch.
Haiku 5.5 Pricing for Developers
Claude Haiku 5.5 costs $0.10 per million input tokens and $0.50 per million output tokens on the Claude Platform for prompts up to 100,000 tokens. Above that, the rate rises five times, to $0.50 and $2.50. Anthropic says the cheaper band is 90% below Haiku 4.5's $1 and $5, the expensive band is 50% below it, and 90% of Haiku 4.5 requests fit under 100,000 tokens. Prompt caching cuts repeat input to $0.01 per million tokens, and the Batch API halves both rates.
Two things narrow the saving. Haiku 5.5 uses a newer tokenizer, so Anthropic's docs say the same text counts as about 30% more tokens than on Haiku 4.5. Simon Willison measured about 1.25 times as many tokens on one long prompt and called it "a hidden price increase". Anthropic's 75% figure already accounts for this. The second is the 100,000-token step: below it Haiku 5.5 matches GPT-6 Luna's price exactly, and above it Luna is cheaper. Our explainer on what AI tokens are shows how to estimate a bill.
Anthropic also launched two related changes the same day. Sonnet 5.5 cache reads dropped from $0.20 to $0.10 per million tokens, which Anthropic says makes Sonnet 5.5 about 20% cheaper on most agentic work. The second is monthly API credits for Max and Team subscribers, rolling out over the week and worth $100 on Max 5x, $200 on Max 20x and up to $500 pooled on Team. The full details are in Anthropic's launch post.
Claude Haiku 5.5 vs Sonnet 5.5 vs GPT-6 Luna: Which to Use
Use Claude Haiku 5.5 for quick jobs on text you give it, and Claude Sonnet 5.5 for anything that needs judgement, long reasoning or code. Anthropic's model guide sorts tasks the same way: Haiku for "simple, straightforward questions with short answers", lookups and extraction, and Sonnet as "your versatile default" for coding, writing and analysis.
| Task | Pick | Why |
|---|---|---|
| Summarise a document or email thread | Haiku 5.5 | Fast, light on your limit, low hallucination rate on supplied text |
| Rewrite, shorten or fix grammar | Haiku 5.5 | Short, well-defined task |
| Pull names, dates or figures out of text | Haiku 5.5 | Extraction is the job Anthropic built it for |
| Write a long piece or a sensitive message | Sonnet 5.5 | Better judgement and tone |
| Write or debug code | Sonnet 5.5 or Opus 5.5 | Anthropic's Terminal-Bench 4.0: 70.6% for Sonnet 5.5 vs 39.2% for Haiku 5.5 |
| Answer a factual question from memory | Sonnet 5.5, or any model with web search | Haiku 5.5 scored 36% on factual recall |
Against GPT-6 Luna, the choice depends on how you use it. In the API the two cost the same up to 100,000 tokens. Haiku 5.5 scores higher on every shared test in Anthropic's table and on the Artificial Analysis index. Luna is cheaper on very long prompts. Our breakdown of GPT-6 Sol and Luna covers where OpenAI ships Luna. For the other side of the Claude lineup, read our Claude Sonnet 5.5 guide.
The Verdict on Claude Haiku 5.5
Claude Haiku 5.5 turns Haiku from a model you avoided into one worth picking on purpose. A year ago Haiku 4.5 scored 17 on the Artificial Analysis index; Haiku 5.5 scores 43, ahead of GPT-6 Luna and Gemini 3.8 Flash. For developers it is a price cut of up to 90% with a catch above 100,000 tokens. For a Claude user, the case is simpler. Switch to Haiku 5.5 for summaries, rewrites and quick extraction on text you supply. It is the lightest model on your usage limit, and it is less likely to make things up. Switch back to Sonnet 5.5 for writing, code and anything you need it to know from memory.
Frequently Asked Questions
When was Claude Haiku 5.5 released?
Anthropic released Claude Haiku 5.5 on October 7, 2026. It is the third Claude 5.5 model, after Opus 5.5 and Sonnet 5.5 in September, and it replaces Claude Haiku 4.5 from October 15, 2025.
Is Claude Haiku 5.5 free?
Yes, inside Claude. Anthropic says Free, Pro, Max, Team and Enterprise users can select Haiku 5.5 on Claude.ai on the web, iOS and Android. Using it through the API costs $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens.
Is Haiku 5.5 better than Sonnet 5.5?
No. Sonnet 5.5 beats Haiku 5.5 on every benchmark Anthropic published and scores 56 to Haiku's 43 on the Artificial Analysis Intelligence Index. Haiku 5.5 is faster, cheaper and lighter on your usage limit, which makes it the better pick for short, well-defined tasks.
What is the Claude Haiku 5.5 context window?
Haiku 5.5 has a 1 million token context window and returns up to 128,000 output tokens, up from 200,000 and 64,000 on Haiku 4.5. Prompts over 100,000 tokens cost five times more per token in the API.
Is Haiku 5.5 better than GPT-6 Luna?
On benchmarks, yes. Haiku 5.5 scores 43 on the Artificial Analysis Intelligence Index to GPT-6 Luna's 38, and it beats Luna on every test in Anthropic's launch table. The two cost the same in the API up to 100,000 tokens, but Luna is cheaper on longer prompts and Haiku 5.5 uses more tokens per task at high effort.
